JP2007086404A - Speech synthesizer - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a speech synthesizer which can easily perform a dictionary registration. <P>SOLUTION: A speech synthesis engine 21 refers to a dictionary database, producing a reading group from a text data representing a group of words and phrases, converting the reading group into speech, outputting the speech from a loudspeaker 25. A candidate selection part 41 selects a portion corresponding to different reading from the text data based on correct reading input from a microphone 35. A speech synthesis user dictionary registration part 42 registers the portion among two or more words and phrases in the dictionary database making them correspond to the correct reading. Thus, when the above text data is given to the speech synthesizer 1, the the speech synthesis engine 21 can correctly read out the above text data by referring to the dictionary database. Using the speech synthesizer 1, a user has only to input correct reading to the microphone 35 when the text data is read out erroneously. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech synthesizer that performs speech synthesis processing.

  Devices for executing speech synthesis processing have been developed. The speech synthesis (Text To Speech) process is a process of converting text data in which a sentence is described into speech and outputting the speech from a speaker. Such speech synthesis processing is very useful when converting non-unique text data into speech, such as reading out an e-mail. However, in speech synthesis processing, there are many cases where reading out text data is wrong. Therefore, a speech synthesizer that corrects different readings of text data has been developed.

  Consider a speech synthesizer that performs speech synthesis processing. The speech synthesizer is provided in a computer and includes a speech synthesis engine, a dictionary database, and a speaker. The dictionary database includes a main dictionary database in which a plurality of words are registered and a user dictionary database. Each of the plurality of words includes a plurality of phrases, a plurality of readings, and a plurality of accents.

  When text data representing a phrase group is given to the speech synthesizer, the speech synthesis engine reads the phrase group represented by the text data. That is, the speech synthesis engine refers to the dictionary database and generates a reading group from the phrase group represented by the text data. Further, the speech synthesis engine determines a pronunciation accent corresponding to the reading group among a plurality of accents. The speech synthesis engine converts the reading group into speech, and outputs the speech from the speaker using the pronunciation accent. Here, it is assumed that reading of text data is wrong. In this case, the user performs dictionary registration using the input device in order to read the text data correctly. That is, the user inputs words / phrases corresponding to different readings of text data, correct readings, and accents using an input device (keyboard, pointing device) and registers them in the user dictionary database. Thus, when the above text data is given to the speech synthesizer, the speech synthesis engine can read the text data correctly.

  However, such a speech synthesizer cannot easily register a dictionary. For example, it is assumed that the user is unfamiliar with the computer. In this case, since the input device is used when the user performs dictionary registration, the operation becomes complicated. For example, it is assumed that the user has difficulty in eyes and hands. In this case, it is difficult for the user to perform dictionary registration using the input device.

  As a technique related to speech synthesis processing, Japanese Patent Laid-Open No. 5-143737 discloses a prosodic symbol generation apparatus (see Patent Document 1). The prosodic symbol generation device is a device that generates a prosodic symbol based on the analysis result of a character string, and includes unregistered word detection means, unregistered word presentation means, and word registration means. Yes. The unregistered word detection means checks whether there is a word that is not registered in the dictionary in the character string to be analyzed when generating the prosodic symbol, and if there is a word that is not registered in the dictionary, this is unregistered Detect as a word. The unregistered word presenting means presents the unregistered word to the operator when the unregistered word is detected. The word registration means enables a predetermined word to be registered in the dictionary prior to character string analysis for generating prosodic symbols. In this prosodic symbol generation device, when generating a reading group from text data representing a phrase group, if the text data includes unregistered words (phrases) that are not registered in the dictionary, the user is prompted to enter a dictionary registration. Prompts.

  Japanese Patent Laid-Open No. 5-233625 discloses a text-to-speech device (see Patent Document 2). The text-to-speech device includes a speech synthesizer, a processing device, a storage device storing a dictionary composed of written character strings, parts of speech, pronunciations, accents, and the like, and an input device for inputting arbitrary text, the dictionary This is an apparatus that obtains word information including pronunciation and accent of a word constituting a sentence from a sentence inputted by reference, and performs voice output of the sentence based on the word information. The text-to-speech device is characterized in that the processing device is provided with first means and second means. The first means estimates the accent from the word information when the accent included in the word information takes a predetermined value indicating that the accent is not registered in the dictionary, and the estimated Sentences are output based on accent and word information. The second means inputs the range designation information indicating the words constituting the inputted sentence and the accent when outputting the words indicated by the range designation information from the input device, and constitutes the words When the word information of the word has the accent unregistered value as an accent, the accent of the word whose accent is unregistered is restored from the input accent and the word information of the word constituting the phrase, and the restoration is performed. The registered accent is registered in the dictionary instead of the accent unregistered value. In this text-to-speech device, dictionary registration in an accent is easily performed by estimating the accent of an unaccented word.

Japanese Patent Laid-Open No. 5-143737 JP-A-5-233625

An object of the present invention is to provide a speech synthesizer that can easily perform dictionary registration.
Another object of the present invention is to provide a speech synthesizer capable of correctly reading text data.

  Hereinafter, means for solving the problem will be described using the numbers and symbols used in [Best Mode for Carrying Out the Invention]. These numbers and symbols are added to clarify the correspondence between the description of [Claims] and the description of [Best Mode for Carrying Out the Invention]. It should not be used to interpret the technical scope of the invention described in “

The speech synthesizer (1) of the present invention is applied to a computer that can use e-mail and the Internet, for example. The computer includes a storage unit (not shown), a control unit (not shown), a speaker (25), and a microphone (35). The storage unit stores a dictionary database that associates a plurality of words with a plurality of readings, and a computer program. The control unit executes the computer program.
The control unit includes a speech synthesis engine (21), a selection unit (41), and a user dictionary registration unit (42).
The speech synthesis engine (21) refers to the dictionary database, generates a reading group from text data representing a phrase group, converts the reading group into speech, and outputs the speech from a speaker (25).
The selection unit (41) selects a part corresponding to a different reading from the text data based on a correct reading input from the microphone (35).
The user dictionary registration unit (42) registers the portion of the plurality of words in the dictionary database in association with the correct reading.
Thus, when the text data is given to the speech synthesizer (1) of the present invention, the speech synthesizer engine (21) refers to the dictionary database {main dictionary database (26), user dictionary database (27)}. Thus, the phrase group represented by the text data can be read out correctly.
Further, according to the speech synthesizer (1) of the present invention, when the reading of the text data is wrong, the user inputs the correct reading to the part corresponding to the different reading of the text data by the microphone (35). Just do it. For this reason, the speech synthesizer (1) of the present invention can perform dictionary registration more easily than when dictionary registration is performed using an input device. Therefore, even a user unfamiliar with a computer can easily register a dictionary. In addition, even a user with impaired eyes and hands can easily register a dictionary.

The speech synthesizer (1) of the present invention further includes a speech recognition engine (31). The voice recognition engine (31) inputs a user's voice from the microphone (35) as the correct reading, and generates a plurality of reading candidates from the user's voice.
The speech synthesis engine (21) generates a plurality of error candidates from the phrase group represented by the text data. The plurality of error candidates include a phrase having a plurality of readings and a phrase that is not registered in the dictionary database.
The selection unit (41) selects an error candidate corresponding to one of the plurality of reading candidates from the plurality of error candidates as the portion.
As described above, in the speech synthesizer (1) of the present invention, when the user inputs a correct reading for the different readings of the text data with the microphone (35), the part corresponding to the different readings of the text data is automatically set. Select and automatically register the dictionary for that part. For this reason, the speech synthesizer (1) of the present invention can perform dictionary registration more easily than when dictionary registration is performed using an input device.

In the speech synthesizer (1) of the present invention, the dictionary database includes a main dictionary database (26) in which the plurality of words, the plurality of readings, and a plurality of accents are registered in association with each other, and a user dictionary database (27). ).
The speech synthesis engine (21) outputs the speech from a speaker (25) using a pronunciation accent corresponding to the reading group among the plurality of accents.
The user dictionary registration unit (42) registers the part of the plurality of words, the correct reading, and the accent by the user's voice in the user dictionary database (27) in association with each other.
As described above, in the speech synthesizer (1) of the present invention, when the user inputs a correct reading for the different readings of the text data with the microphone (35), the portion corresponding to the different readings of the text data and the correct reading And the user's voice accent are associated and registered in the dictionary. Usually, it is not so easy to register a dictionary for an accent using an input device due to the user's own habit and dialect. In the speech synthesizer (1) according to the present invention, since the accent is directly registered in the dictionary by the user's voice, the dictionary can be registered more easily than when the dictionary is registered using the input device.
In the speech synthesizer (1) of the present invention, essential words (phrases, readings, accents) are registered in the dictionary in advance in the main dictionary database (26), and words that are not registered in the main dictionary database (26). It is preferable to register the above part of the plurality of words, the correct reading, and the accent by the user's voice in the user dictionary database (27). Thus, when the text data is given to the speech synthesizer (1) of the present invention, the speech synthesizer engine (21) refers to the dictionary database {main dictionary database (26), user dictionary database (27)}. Thus, the phrase group represented by the text data can be read out correctly.

In the speech synthesizer (1) of the present invention, the dictionary database further includes a category-added user dictionary database (54).
The text data includes information representing a category.
When a category registration instruction is generated, the following operation is performed.
In this case, the user dictionary registration unit (42) associates the category, the portion of the plurality of words / phrases, the correct reading, and the accent by the user's voice with the category-added user dictionary database. Register at (54).
For example, when the text data is a web page representing an article, “sports”, “politics”, “music”,... Can be acquired relatively easily as information representing the category described in the article. In the speech synthesizer (1) of the present invention, when information representing a category is included in the text data, the category, a word that is not registered in the main dictionary database (26) (the portion of the plurality of words and phrases, It is preferable to register the dictionary in the category-added user dictionary database (54) by associating the correct reading with the accent by the user's voice. As a result, when the above text data is given to the speech synthesizer (1) of the present invention, the speech synthesis engine (21) uses the dictionary database {main dictionary database (26), category-added user dictionary database (54)}. Referring to the above, it is possible to correctly read out the phrase group represented by the text data. For this reason, the speech synthesizer (1) of the present invention can further reduce reading errors as compared with the case of referring to the user dictionary database (27).

In the speech synthesizer (1) of the present invention, the dictionary database further includes a user dictionary database with co-occurrence information (64).
When a co-occurrence information registration instruction is generated, the following operation is performed.
In this case, the speech recognition engine (31) inputs a connected voice that is a user's voice from the microphone (35) as a reading of a connected portion connected to the portion of the text data, and a plurality of connected voices are connected from the connected speech. Generate reading candidates.
The speech synthesis engine (21) generates a plurality of co-occurrence candidates corresponding to the connected portion from the text data.
The selection unit (41) selects a co-occurrence candidate corresponding to one of the plurality of linked reading candidates from the plurality of co-occurrence candidates as the linked portion.
The user dictionary registration unit (42) registers the first word, the second word, and the co-occurrence information that associates the first word with the second word in the user dictionary database with co-occurrence information (64).
The first word includes the portion of the plurality of phrases, the correct reading, and an accent by the user's voice, and the second word includes the connected portion of the plurality of phrases; It includes a reading of the connected part and an accent by the connected voice.
For example, when a proper noun (such as a person's name) is read out by speech synthesis, there are many cases where the same word or phrase is read differently, and it is difficult to distinguish and read it out. When the proper noun is a person name, it is preferable to register the dictionary with the full name. In this case, in the speech synthesizer (1) of the present invention, the first word (the portion of the plurality of phrases, the correct reading, the accent by the user's voice), and the second word (the plurality of phrases). The above-mentioned connected portion, reading of the connected portion, accent by the connected speech), and co-occurrence information that associates the first word with the second word are registered in the user dictionary database with co-occurrence information (64). It is preferable to do. As a result, when the above text data is given to the speech synthesizer (1) of the present invention, the speech synthesis engine (21) reads the dictionary database {main dictionary database (26), user dictionary database with co-occurrence information (64). )}, The group of words represented by the text data can be read correctly. For this reason, the speech synthesizer (1) of the present invention can further reduce reading errors as compared with the case of referring to the user dictionary database (27).

The speech synthesizer of the present invention can easily perform dictionary registration.
The speech synthesizer of the present invention can read text data correctly.

  The speech synthesizer of the present invention will be described below in detail with reference to the accompanying drawings.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of a speech synthesis apparatus according to the first embodiment of the present invention. The speech synthesizer 1 according to the first embodiment is applied to a computer that can use e-mail or the Internet, for example. The computer includes a storage unit (not shown) that stores a computer program and a control unit (not shown) that executes the computer program.
The voice synthesizer 1 includes a speaker 25 and a microphone 35 provided in a computer.

The speech synthesizer 1 further includes a sentence extraction unit 12, which is a computer program, a speech synthesis engine 21, a speech recognition engine 31, a candidate selection unit 41, and a synthesized speech user dictionary registration unit 42.
The speech synthesis engine 21 includes a speech synthesis unit 22, a syntax analysis unit 23, and an error candidate generation unit 24. The voice synthesizer 22 is a main body of the voice synthesizer 21, and the voice synthesizer 22 may be referred to as the voice synthesizer 21.
The speech recognition engine 31 includes a speech recognition unit 32, a reading candidate generation unit 33, and an accent extraction unit 34. The voice recognition unit 32 is a main body of the voice recognition engine 31, and the voice recognition unit 32 may be referred to as the voice recognition engine 31.

The speech synthesizer 1 further includes a document database 11, a speech synthesis dictionary database, and a speech recognition dictionary database 36 stored in the storage unit of the computer.
The speech synthesis dictionary database includes a speech synthesis main dictionary database 26 and a speech synthesis user dictionary database 27.
For example, as shown in FIG. 2, a plurality of words are registered in advance in the speech synthesis main dictionary database 26. Each of the plurality of words includes a plurality of phrases, a plurality of readings, and a plurality of accents. As shown in FIG. 3, words that are not registered in the speech synthesis main dictionary database 26 are registered in the speech synthesis user dictionary database 27.
As shown in FIG. 4, a plurality of collation voices and a plurality of collation readings are registered in advance in the speech recognition dictionary database 36. The plurality of collation voices are data for collating voices input from the microphone 35.

  As shown in FIG. 5, the document database 11 stores text data representing a document exemplified by an e-mail or an Internet web page. For example, in the text data, there is a sentence “Hideki Okada has scored at the end of the seventh.”

  The sentence extraction unit 12 includes a built-in memory (not shown). In response to the user's reading instruction, the sentence extraction unit 12 reads the text data stored in the document database 11 and stores it in the built-in memory. Alternatively, when the computer acquires text data representing an e-mail or a web page and stores it in the document database 11, the sentence extraction unit 12 automatically reads out the text data stored in the document database 11, and stores the internal memory. To store.

  The syntax analysis unit 23 of the speech synthesis engine 21 reads the text data from the built-in memory of the sentence extraction unit 12 and analyzes the sentence described in the text data. At this time, the syntax analysis unit 23 uses the kanji, hiragana, katakana, and numbers (including numbers representing the number of times and quantity) as text data “Hideki Okada scored at the end of the 7th”. It is recognized that “7 times”, “back”, “ni”, “Okada”, “Hideki”, “has”, “RBI”, “I” and “I gave”.

  The syntax analysis unit 23 refers to the speech synthesis dictionary database (the speech synthesis main dictionary database 26, the speech synthesis user dictionary database 27), and the phrase group represented by the text data “Hideki Okada gave a hit point on the back of the 7th.” From this, a group of readings representing readings is generated, "Nakanaka Hideyoshi was raised". In addition, the syntax analysis unit 23 refers to the speech synthesis dictionary database (speech synthesis main dictionary database 26, speech synthesis user dictionary database 27), and determines a pronunciation accent corresponding to the reading group among a plurality of accents. To do.

  The speech synthesizer 22 of the speech synthesis engine 21 converts the reading represented by the reading group “Nakanaka Hideyoshi was raised” into speech. The voice synthesizer 22 outputs the voice from the speaker 25 using the above-mentioned pronunciation accent.

  The original reading for the text data “Hideki Okada scored at the back of the seventh episode” is “I've given you the best results”. That is, the reading for “Hideki” is not “Hideyoshi” but “Hide”. The user inputs the user's voice “Hide” using the microphone 35 in order to correct the portion “Hideki” corresponding to the different reading “Hideyoshi” of the text data to the correct reading “Hide”.

  The voice recognition unit 32 of the voice recognition engine 31 inputs the user's voice “hidori” from the microphone 35 as correct reading “hidori”. The candidate selection unit 41 selects a portion “Hideki” corresponding to a different reading “Hideyoshi” from the text data 80 based on the correct reading “Hide” input from the microphone 35. This will be specifically described.

  The voice recognition engine 31 further includes a built-in memory (not shown). The voice recognition unit 32 of the voice recognition engine 31 inputs the user's voice “hidori” from the microphone 35 and stores it in the built-in memory as data. The reading candidate generation unit 33 of the voice recognition engine 31 refers to the voice recognition dictionary database 36 and generates a plurality of reading candidates from the user's voice “hidori” stored in the built-in memory. That is, among a plurality of collation voices, a plurality of reading candidates “hidashi” “hidashi” “hidashi” “iideshi”... Are generated in the order closest to the user's voice “hidori”.

  The error candidate generation unit 24 of the speech synthesis engine 21 uses the above word groups “7 times”, “back”, “ni”, “Okada”, “Hideki” represented by the text data “Hideki Okada scored at the back of the seventh time”. A plurality of error candidates are generated from “G”, “RBI”, “Wa” and “Raised”. In other words, the candidate selection unit 41 uses a phrase “with 7”, “back”, “ni”, “Okada”, “Hideki”, “ga”, “batting point”, “a”, “raised”, and a phrase “with multiple readings”. “Hideki”, “RBI”, and words that are not registered in the speech synthesis dictionary databases 26 and 27 are selected as error candidates.

  The candidate selection unit 41 compares a plurality of error candidates “Hideki”, “dots”,... With a plurality of reading candidates “Hide”, “Hideshi”, “Ise”, “Ideshi”,. As a result of the comparison, the candidate selection unit 41 selects one of a plurality of reading candidates “Hide”, “Hideshi”, “Idei”, “Ideshi”,... The error candidate “Hideki” corresponding to the reading candidate “Hide” is selected. That is, the candidate selection unit 41 refers to the speech synthesis dictionary databases 26 and 27 described above, and converts the reading candidate “Hide” into the kanji characters “Hideki”, “Hideki”, “Hideki”, “Hideki”, “Hideki”,. . The candidate selection unit 41 corresponds to (matches) one of the kanji characters “Hideki”, “Hideki”, “Hideki”, “Hideki”, “Takashi Hiji”, etc. Select the candidate “Hideki”. In this way, the error candidate “Hideki” corresponding to the correct reading “Hide” is selected from the text data. The selected error candidate “Hideyoshi” corresponds to a different reading “Hideyoshi”.

As a result of the comparison, the candidate selection unit 41 reads out the phrase “Hideki”, which is the part of the plurality of phrases stored in the speech synthesis main dictionary database 26, and reads the phrase “Hideki” and the correct reading “Hideki”. To the speech synthesis user dictionary registration unit 42.
At the same time, the accent extraction unit 34 of the speech recognition engine 31 reads the user's voice “hidori” stored in the built-in memory, extracts the accent from the voice “hidori”, and notifies the voice synthesis user dictionary registration unit 42 of the accent.
The speech synthesis user dictionary registration unit 42 registers the phrase “Hideki”, the correct reading “hidori”, and the accent in the speech synthesis user dictionary database 27 in association with each other.

  Thereby, when the text data is stored in the document database 11, the speech synthesis engine 21 can correctly read out the word / phrase group represented by the text data. That is, the speech synthesis engine 21 refers to the speech synthesis dictionary database (speech synthesis main dictionary database 26, speech synthesis user dictionary database 27), and the word group “Hideki Okada scored the back of the 7th phrase represented by the text data. ”Is generated from the reading group“ Naka Nakadaka Gadgeten was raised ”. In addition, the accent for pronunciation corresponding to the reading group is determined. The speech synthesizing engine 21 converts the reading group “I have raised my name” to speech and outputs the speech from the speaker 25 using the above-mentioned accents for pronunciation.

Next, the operation (speech synthesis method / speech synthesis process) of the speech synthesizer 1 according to the first embodiment of the present invention will be described with reference to FIG.
Here, the document database 11 stores text data representing an Internet Web page. In this text data, a plurality of sentences are described. In addition, the speech synthesizer 1 converts a sentence described in the text data into speech when the computer acquires the text data and stores it in the document database 11.

  The sentence extraction unit 12 reads the text data stored in the document database 11 and stores it in the built-in memory of the sentence extraction unit 12 (step S1).

The syntax analysis unit 23 of the speech synthesis engine 21 recognizes a plurality of sentences based on sentence delimiters such as “.” And line breaks written in the text data from the built-in memory of the sentence extraction unit 12. The syntax analysis unit 23 sets the variable N to 1 (N = 1), and reads the first sentence among the plurality of sentences (step S2).
The syntax analysis unit 23 analyzes the sentence. At this time, the syntax analysis unit 23 recognizes a word / phrase group represented by the sentence based on the kanji, hiragana, katakana, and numbers (step S3). The syntax analysis unit 23 refers to the speech synthesis main dictionary database 26 and the speech synthesis user dictionary database 27 (steps S4 and S5). At this time, the syntax analysis unit 23 generates a reading group from the phrase group represented by the sentence, and determines a pronunciation accent corresponding to the reading group (step S6).

  The speech synthesizer 22 of the speech synthesis engine 21 reads out the phrase group represented by the sentence. That is, the readings represented by the reading group are converted into speech, and the speech is output from the speaker 25 using the above-mentioned accent for pronunciation (step S7). If the user does not use the microphone 35 while the voice synthesizer 22 outputs the above voice from the speaker 25 (steps S8-NO and S9), the syntax analyzer 23 increments the variable N (N = N + 1), and the next sentence is read (step S10-YES, S11). If there is no next sentence (step S10—NO), the speech synthesizer 1 ends the speech synthesis process.

  On the other hand, the user uses the microphone 35 while the voice synthesizer 22 outputs the above voice from the speaker 25 (step S8—YES). That is, the user inputs the user's voice using the microphone 35 in order to correct a part corresponding to a different reading from the above sentence into a correct reading. At this time, the voice recognition unit 32 of the voice recognition engine 31 inputs the user's voice from the microphone 35 and stores it in the built-in memory of the voice recognition engine 31 (step S12). The reading candidate generation unit 33 of the voice recognition engine 31 refers to the voice recognition dictionary database 36 and generates a plurality of reading candidates from the user's voice stored in the built-in memory (step S13).

  The speech synthesizer 22 of the speech synthesizer 21 sets the variable I to 1 (I = 1), and reads the sentence again from the built-in memory of the sentence extractor 12 (step S14). The syntax analysis unit 23 of the speech synthesis engine 21 analyzes the sentence and recognizes a phrase group represented by the sentence (step S15). The error candidate generation unit 24 of the speech synthesis engine 21 includes a plurality of error candidates including a phrase having a plurality of readings and a phrase not registered in the speech synthesis dictionary databases 26 and 27 from the phrase group represented by the sentence. Is generated (step S16). The candidate selection unit 41 compares a plurality of error candidates with a plurality of reading candidates (step S17).

As a result of the comparison, the candidate selection unit 41 cannot select an error candidate corresponding to one reading candidate among the plurality of reading candidates from the plurality of error candidates (step S18—NO).
At this time, when the variable I is smaller than the variable N (step S19—NO), the speech synthesizer 1 executes steps S15 to S19 for the previous sentence.
If the variable I is the variable N, or if the variable I is smaller than the variable N but there is no previous sentence (step S19-YES), the sentence read out in step S14 is used as the sentence extraction unit. Extract from 12 internal memories. That is, the speech synthesis unit 22 of the speech synthesis engine 21 reads the sentence again from the built-in memory of the sentence extraction unit 12 (step S23). Thereafter, the speech synthesizer 1 executes the above step S3 and subsequent steps for the sentence.

On the other hand, as a result of the comparison, the candidate selection unit 41 selects an error candidate corresponding to one reading candidate among the plurality of reading candidates from the plurality of error candidates. At this time, as a result of the comparison, the candidate selection unit 41 reads out the phrase that is the error candidate among the plurality of phrases stored in the speech synthesis main dictionary database 26, and reads the read phrase and the correct reading. Is notified to the speech synthesis user dictionary registration unit 42 (YES in step S18).
At the same time, the accent extraction unit 34 of the speech recognition engine 31 reads the user's voice stored in the built-in memory, extracts the accent from the voice, and notifies the voice synthesis user dictionary registration unit 42 (step S21).
The speech synthesis user dictionary registration unit 42 registers the phrase, correct reading notified from the candidate selection unit 41, and the accent notified from the accent extraction unit 34 in association with each other in the speech synthesis user dictionary database 27 (step S22). .
Thereafter, the speech synthesizer 1 executes the above step S23 and subsequent steps for the sentence.

As described above, in the speech synthesizer 1 of the present invention, the speech synthesis engine 21 refers to the speech synthesis dictionary database (speech synthesis main dictionary database 26, speech synthesis user dictionary database 27), and the phrase group represented by the text data. A reading group is generated from the voice, the reading group is converted into voice, and the voice is output from the speaker 25. The candidate selection unit 41 selects a part corresponding to a different reading from the text data based on the correct reading input from the microphone 35. The speech synthesis user dictionary registration unit 42 registers the part of the plurality of words in the speech synthesis user dictionary database 27 in association with the correct reading. Thus, when the text data is given to the speech synthesizer 1 of the present invention, the speech synthesis engine 21 refers to the speech synthesis dictionary database (speech synthesis main dictionary database 26, speech synthesis user dictionary database 27). The phrase group represented by the text data can be read correctly.
Also, according to the speech synthesizer 1 of the present invention, when the text data is read out incorrectly, the user only has to input the correct reading to the part corresponding to the different text data reading by the microphone 35. . For this reason, the speech synthesizer 1 of the present invention can easily perform dictionary registration as compared to the case of dictionary registration using an input device. Therefore, even a user unfamiliar with a computer can easily register a dictionary. In addition, even a user with impaired eyes and hands can easily register a dictionary.

  As described above, in the speech synthesizer 1 of the present invention, the speech recognition engine 31 inputs the user's voice from the microphone 35 as the correct reading and generates a plurality of reading candidates from the user's voice. The speech synthesis engine 21 generates a plurality of error candidates from the text data. The candidate selection unit 41 selects an error candidate corresponding to one of a plurality of reading candidates as a part from among a plurality of error candidates. Thus, in the speech synthesizer 1 of the present invention, when the user inputs a correct reading with the microphone 35 for different readings of the text data, the part corresponding to the different readings of the text data is automatically selected, The dictionary is automatically registered for the part. For this reason, the speech synthesizer 1 of the present invention can easily perform dictionary registration as compared to the case of dictionary registration using an input device.

As described above, in the speech synthesizer 1 of the present invention, the speech synthesis engine 21 outputs the speech from the speaker 25 using the accent for pronunciation corresponding to the reading group among a plurality of accents. The speech synthesis user dictionary registration unit 42 registers the portion of the plurality of words, the correct reading, and the accent by the user's speech in the speech synthesis user dictionary database 27 in association with each other. As described above, in the speech synthesizer 1 according to the present invention, when the user inputs a correct reading for the different readings of the text data with the microphone 35, the portion corresponding to the different readings of the text data, the correct reading, The dictionary is registered in association with the accent by voice. Usually, it is not so easy to register a dictionary for an accent using an input device due to the user's own habit and dialect. In the speech synthesizer 1 according to the present invention, since the accent is directly registered in the dictionary by the user's voice, the dictionary can be registered more easily than in the case where the dictionary is registered using the input device.
In the speech synthesizer 1 of the present invention, essential words (phrases, readings, accents) are registered in the dictionary in advance in the speech synthesis main dictionary database 26, and words (a plurality of words not registered in the speech synthesis main dictionary database 26) are registered. It is preferable to register the above part of the phrase, the above correct reading, and the accent by the user's voice) in the speech synthesis user dictionary database 27. Thus, when the text data is given to the speech synthesizer 1 of the present invention, the speech synthesis engine 21 refers to the speech synthesis dictionary database (speech synthesis main dictionary database 26, speech synthesis user dictionary database 27). The phrase group represented by the text data can be read correctly.

(Second Embodiment)
A speech synthesizer 1 according to a second embodiment of the present invention will be described. In the second embodiment, only differences from the first embodiment will be described. In the second embodiment, a category-added dictionary registration process and a co-occurrence dictionary registration process are realized.

  In the dictionary registration process with categories, when reading proper nouns (person names, etc.) by speech synthesis, there are many cases where the same words and phrases are read differently, and it is difficult to distinguish them and read them out. If the document to be read is, for example, a Web page in which the text data represents an article, “sports”, “politics”, “music”,... Can be acquired relatively easily as information representing the category described in the article. . In this case, the speech synthesizer 1 is provided with a category-added dictionary database as one of the speech synthesis dictionary databases, and the category-added dictionary is created by associating categories with words including words, phrases, correct readings, and accents when registering the dictionary. By registering with, reading errors can be further reduced.

  In the co-occurrence dictionary registration process, for example, when a proper noun (person name, etc.) is read out by speech synthesis, there are many cases where the same word / phrase is read differently, and it is difficult to distinguish and read it out. When the proper noun is a person name, it is preferable to register the dictionary with the full name. In this case, the speech synthesizer 1 is provided with a dictionary with co-occurrence information as one of the speech synthesis dictionary databases, and at the time of dictionary registration, co-occurrence information that combines the phrase and the linked phrase linked to the phrase, To further reduce reading errors by associating words including correct readings and accents, concatenated phrases (co-occurrence words), words including the readings and accents, and registering them in the dictionary with co-occurrence information Can do.

First, registration in the dictionary with categories will be described.
As shown in FIG. 7, the speech synthesizer 1 further includes a category registration trigger switch 51 provided in the computer.
The speech synthesizer 1 further includes a category extraction unit 52 that is a computer program.
The speech synthesizer 1 further includes a category extraction information database 53 stored in the storage unit of the computer. In the category extraction information database 53, a plurality of reference categories “sports”, “politics”, “music”,... Are registered.
The above speech synthesis dictionary database stored in the storage unit of the computer further includes a user dictionary database 54 with a speech synthesis category.

In the text data, information representing the category “sports” is described, and a sentence “Hideki Okada has scored a hit on the 7th time” is described. Consider the case where the speech synthesizer 22 of the speech synthesizer engine 21 outputs the speech “Nakadaka Hideyoshi Gaden” from the speaker 25 for this text data.
In this case, the user inputs the user's voice “Hide” using the microphone 35 in order to correct the portion “Hideyoshi” corresponding to the different reading “Hideyoshi” of the text data to the correct reading “Hide”. At this time, the user operates the category registration trigger switch 51 in order to associate the word “Hideki”, which is the above part, the word including the correct reading “hidori” and the accent with the category.

  The category extraction unit 52 notifies the speech synthesis user dictionary registration unit 42 of a category registration instruction indicating that the category registration trigger switch 51 has been operated. As shown in FIG. 8, first, the speech synthesis user dictionary registration unit 42, in response to the category registration instruction, converts a word including the phrase “Hideki”, a correct reading “Hide”, and an accent into a user dictionary with a speech synthesis category. Register in the database 54. The category extraction unit 52 reads the text data from the built-in memory of the sentence extraction unit 12 and extracts the category “sports” described in the text data. That is, referring to the category extraction information database 53, the reference category “sports” that matches the category described in the text data is selected from the plurality of reference categories “sports” “politics” “music”. . The category extraction unit 52 notifies the speech synthesis user dictionary registration unit 42 of the category “sports”. As shown in FIG. 8, the speech synthesis user dictionary registration unit 42 responds to the category registration instruction with the category “sports” including the word “Hideki”, the correct reading “hidori”, and the accent. Are registered in the user dictionary database 54 with a speech synthesis category in association with

Next, registration in the co-occurrence dictionary will be described.
As shown in FIG. 7, the speech synthesizer 1 further includes a co-occurrence information registration trigger switch 61 provided in the computer.
The speech synthesizer 1 further includes a co-occurrence information extraction unit 62 that is a computer program.
The above speech synthesis dictionary database stored in the storage unit of the computer further includes a user dictionary database 64 with speech synthesis co-occurrence information.

In the text data, information representing the category “sports” is described, and a sentence “Hideki Okada has scored a hit on the 7th time” is described. Consider the case where the speech synthesizer 22 of the speech synthesizer engine 21 outputs the speech “Nakadaka Hideyoshi Gaden” from the speaker 25 for this text data.
In this case, the user inputs the user's voice “Hide” using the microphone 35 in order to correct the portion “Hideyoshi” corresponding to the different reading “Hideyoshi” of the text data to the correct reading “Hide”. At this time, the user associates the word including the phrase “Hideki”, the correct reading “Hide”, and the accent with the word “Okada”, the reading “Okada”, and the word including the accent. In addition, the co-occurrence information registration trigger switch 61 is operated, and the connected voice “Okada”, which is the voice of the user, is input using the microphone 35.

The co-occurrence information extraction unit 62 notifies the speech synthesis user dictionary registration unit 42 of a co-occurrence information registration instruction indicating that the co-occurrence information registration trigger switch 61 has been operated. As shown in FIG. 9, in response to the co-occurrence information registration instruction, the speech synthesis user dictionary registration unit 42 synthesizes the first word including the word “Hideki”, the correct reading “hidashi”, and the accent. Register in the user dictionary database 64 with co-occurrence information. At the same time, first co-occurrence information to be described later is registered in the user dictionary database 64 with speech synthesis co-occurrence information in association with the first word.
The co-occurrence information extraction unit 62 notifies the co-occurrence information registration instruction to the speech recognition engine 31, the speech synthesis engine 21, and the candidate selection unit 41.

  In response to the co-occurrence information registration instruction, the voice recognition unit 32 of the voice recognition engine 31 reads the above-mentioned connected voice “Okada” from the microphone 35 as a reading of the connected part “Okada” connected to the above-mentioned part “Hideki” of the text data. Input and store it in the built-in memory as data. The reading candidate generation unit 33 of the speech recognition engine 31 refers to the speech recognition dictionary database 36 and generates a plurality of linked reading candidates from the linked speech “okada” stored in the built-in memory. That is, a plurality of linked reading candidates “okada”... Are generated from the plurality of collation voices in the order closest to the linked speech “okada”.

  In response to the co-occurrence information registration instruction, the error candidate generation unit 24 of the speech synthesis engine 21 selects the word groups “seven times”, “back”, and “represented by the text data“ Hideki Okada has scored at the end of 7 times ”. A plurality of co-occurrence candidates corresponding to the above-mentioned connected portions are generated from “Okada”, “Hideki”, “Ga”, “RBI”, “O”, and “Raised”. That is, the candidate selection unit 41 connects the phrase group “seven times”, “back”, “ni”, “Okada”, “Hideki”, “ga”, “batting point”, “raised”, and “raised” to the phrase “Hideki”. The words “Okada” and “ga” are selected as co-occurrence candidates.

  The candidate selection unit 41 compares a plurality of co-occurrence candidates “Okada” “ga” with a plurality of linked reading candidates “okada”... According to the co-occurrence information registration instruction. As a result of the comparison, the candidate selection unit 41 selects a co-occurrence candidate corresponding to one reading candidate “Okada” from among a plurality of linked reading candidates “Okada”... Select “Okada”. That is, the candidate selection unit 41 refers to the above-described speech synthesis dictionary database (speech synthesis main dictionary database 26, speech synthesis user dictionary database 27), and converts the concatenated reading candidate “okada” to the kanji “Okada”. The candidate selection unit 41 selects a co-occurrence candidate “Okada” corresponding to (matching) one of the kanji characters “Okada”... From a plurality of co-occurrence candidates “Okada” and “ga”. In this way, the co-occurrence candidate “Okada” corresponding to the correct reading “hidori” is selected.

As a result of the comparison, the candidate selection unit 41 reads out the concatenated phrase “Okada”, which is a concatenated part connected to the above part, from among a plurality of words stored in the speech synthesis main dictionary database 26, and the concatenated phrase “Okada” And the reading “Okada” to the speech synthesis user dictionary registration unit 42.
At the same time, the accent extraction unit 34 of the speech recognition engine 31 reads the connected speech “Okada” stored in the built-in memory in response to the co-occurrence information registration instruction, extracts the accent from the speech “Okada”, and generates a speech synthesis user. Notify the dictionary registration unit 42.
In response to the co-occurrence information registration instruction, the speech synthesis user dictionary registration unit 42 stores the second word including the concatenated phrase “Okada”, its reading “Okada” and its accent in the user dictionary database 64 with speech synthesis co-occurrence information. sign up. At the same time, second co-occurrence information to be described later is associated with the second word and registered in the user dictionary database 64 with speech synthesis co-occurrence information. The first co-occurrence information and the second co-occurrence information are co-occurrence information that associates the first word with the second word. As described above, the concatenated phrase “Okada” is a concatenated part that is concatenated before the part “Hideki” of the text data. For this reason, the speech synthesis user dictionary registration unit 42 registers the second co-occurrence information and the first co-occurrence information so that the words “Okada” and “Hideki” are registered in the user dictionary database 64 with speech synthesis co-occurrence information in this order. Order information “1-1” and “1-2” indicating the order of word arrangement is added to the origin information.

Next, the operation (speech synthesis method / speech synthesis process) of the speech synthesizer 1 according to the second embodiment of the present invention will be described with reference to FIG.
Here, as in the first embodiment, the document database 11 stores text data representing Internet web pages. In this text data, a plurality of sentences are described. In addition, the speech synthesizer 1 converts a sentence described in the text data into speech when the computer acquires the text data and stores it in the document database 11.

  The operation of the speech synthesizer 1 according to the second embodiment is different in step S22 from the first embodiment. When executing step S22, the speech synthesizer 1 checks whether or not the category registration trigger switch 51 or the co-occurrence information registration trigger switch 61 is operated as a trigger (step S31), and performs user dictionary registration processing, category Either an attached dictionary registration process or a co-occurrence dictionary registration process is executed.

Here, the speech synthesis user dictionary registration unit 42 does not accept a category registration instruction and a co-occurrence information registration instruction. In this case, the speech synthesis user dictionary registration unit 42 recognizes that the category registration trigger switch 51 and the co-occurrence information registration trigger switch 61 are not operated (step S31—no trigger), and the user dictionary registration process is executed. Is done.
In this user dictionary registration process, as in the first embodiment, the speech synthesis user dictionary registration unit 42 associates the word / phrase notified from the candidate selection unit 41, the correct reading, and the accent notified from the accent extraction unit 34. At the same time, it is registered in the speech synthesis user dictionary database 27 (step S32).

The user inputs the user's voice using the microphone 35 (step S8—YES), and operates the category registration trigger switch 51. The category extraction unit 52 outputs a category registration instruction indicating that the category registration trigger switch 51 has been operated, and the speech synthesis user dictionary registration unit 42 receives the category registration instruction. In this case, the speech synthesis user dictionary registration unit 42 recognizes that the category registration trigger switch 51 has been operated, and executes a category-added dictionary registration process.
In this category-added dictionary registration process, the speech synthesis user dictionary registration unit 42 responds to the word / phrase notified from the candidate selection unit 41, the correct reading, and the accent notified from the accent extraction unit 34 according to the category registration instruction. At the same time, it is registered in the speech synthesis user dictionary database 27 (step S31—with category trigger).
The category extraction unit 52 reads the text data from the built-in memory of the sentence extraction unit 12 (step S41), refers to the category extraction information database 53, and extracts the category described in the text data (steps S42 and S43). The category extraction unit 52 notifies the speech synthesis user dictionary registration unit 42 of this category. In response to the category registration instruction, the speech synthesis user dictionary registration unit 42 registers the category in the user dictionary database 54 with the speech synthesis category in association with the word including the word, the correct reading, and the accent. (Step S44).

After inputting the user's voice using the microphone 35 (step S8—YES), the user operates the co-occurrence information registration trigger switch 61 and inputs the connected voice that is the user's voice using the microphone 35. The co-occurrence information extraction unit 62 outputs a co-occurrence information registration instruction indicating that the co-occurrence information registration trigger switch 61 has been operated, and the speech synthesis user dictionary registration unit 42 receives the co-occurrence information registration instruction. In this case, the speech synthesis user dictionary registration unit 42 recognizes that the co-occurrence information registration trigger switch 61 has been operated, and executes the co-occurrence dictionary registration process.
In this co-occurrence dictionary registration process, the speech synthesis user dictionary registration unit 42, in response to the co-occurrence information registration instruction, the phrase notified from the candidate selection unit 41, correct reading, and the accent notified from the accent extraction unit 34. Is registered in the speech synthesis user dictionary database 27 as the first word. At the same time, the speech synthesis user dictionary registration unit 42 registers the first co-occurrence information in the speech synthesis user dictionary database 27 in association with the first word (step S31—with co-occurrence word trigger).

  In response to the co-occurrence information registration instruction, the speech recognition unit 32 of the speech recognition engine 31 inputs the above connected speech from the microphone 35 as a reading of the connected portion connected to the above portion of the sentence described in the text data, Data is stored in the built-in memory (step S51). The reading candidate generation unit 33 of the speech recognition engine 31 refers to the speech recognition dictionary database 36 and generates a plurality of linked reading candidates from the linked speech stored in the built-in memory (step S52).

  In response to the co-occurrence information registration instruction, the error candidate generation unit 24 of the speech synthesis engine 21 generates a plurality of co-occurrence candidates corresponding to the above-described connected portion from the word / phrase group represented by the sentence (step S53).

In response to the co-occurrence information registration instruction, the candidate selecting unit 41 compares the plurality of co-occurrence candidates with the plurality of linked reading candidates (step S54).
As a result of the comparison, the candidate selection unit 41 selects a co-occurrence candidate corresponding to one reading candidate among the plurality of linked reading candidates from the plurality of co-occurrence candidates. At this time, as a result of the comparison, the candidate selection unit 41 reads out a connected word / phrase that is a connected part connected to the above part from among a plurality of words / phrases stored in the speech synthesis main dictionary database 26, and The reading is notified to the speech synthesis user dictionary registration unit 42. At the same time, the accent extraction unit 34 of the speech recognition engine 31 reads the connected speech stored in the built-in memory in response to the co-occurrence information registration instruction, extracts the accent from the speech, and notifies the speech synthesis user dictionary registration unit 42 (Step S55).
In response to the co-occurrence information registration instruction, the speech synthesis user dictionary registration unit 42 selects a word including the connected phrase and the reading notified from the candidate selection unit 41 and the accent notified from the accent extraction unit 34 as the second word. Is registered in the speech synthesis user dictionary database 27. At the same time, the speech synthesis user dictionary registration unit 42 registers the second co-occurrence information in the speech synthesis user dictionary database 27 in association with the second word. At this time, the speech synthesis user dictionary registration unit 42 adds the order information “1-1” and “1-2” indicating the order of the arrangement of the words to the second co-occurrence information and the first co-occurrence information, respectively (step) S56).

  As described above, for example, when the text data is a Web page representing an article, “sports”, “politics”, “music”,... Can be acquired relatively easily as information representing a category described in the article. it can. In the speech synthesizer 1 of the present invention, when information representing a category is included in the text data, the category and the words not registered in the speech synthesis main dictionary database 26 (the portion of the plurality of phrases and the correct reading) It is preferable to register the dictionary in the speech synthesis category-added user dictionary database 54 in association with the user's voice accent). Thus, when the text data is given to the speech synthesizer 1 of the present invention, the speech synthesis engine 21 refers to the speech synthesis dictionary database (the speech synthesis main dictionary database 26, the user dictionary database 54 with the speech synthesis category). Thus, the phrase group represented by the text data can be read out correctly. For this reason, the speech synthesizer 1 of the present invention can further reduce reading errors as compared with the case where the speech synthesis user dictionary database 27 is referred to.

  As described above, for example, when a proper noun (person name, etc.) is read out by speech synthesis, there are many cases where the same word or phrase is read differently, and it is difficult to distinguish and read it out. When the proper noun is a person name, it is preferable to register the dictionary with the full name. In this case, in the speech synthesizer 1 of the present invention, the first word (the portion of the plurality of phrases, the correct reading, the accent by the user's voice), and the second word (the connection of the plurality of phrases). It is preferable to register the part, the reading of the connected part, the accent by the connected voice), and the co-occurrence information associating the first word with the second word in the user dictionary database 64 with co-occurrence information. As a result, when the above text data is given to the speech synthesizer 1 of the present invention, the speech synthesis engine 21 reads the speech synthesis dictionary database (speech synthesis main dictionary database 26, user dictionary database 64 with speech synthesis co-occurrence information). Referring to the above, it is possible to correctly read out the phrase group represented by the text data. For this reason, the speech synthesizer 1 of the present invention can further reduce reading errors as compared with the case where the speech synthesis user dictionary database 27 is referred to.

FIG. 1 is a block diagram showing the configuration of the speech synthesizer 1 of the present invention. (First embodiment) FIG. 2 shows the speech synthesis main dictionary database 26 of the speech synthesizer 1 of the present invention. (First embodiment) FIG. 3 shows the speech synthesis user dictionary database 27 of the speech synthesizer 1 of the present invention. (First embodiment) FIG. 4 shows the speech recognition dictionary database 36 of the speech synthesizer 1 of the present invention. (First embodiment) FIG. 5 is a diagram for explaining the operation of the speech synthesizer 1 of the present invention. (First embodiment, second embodiment) FIG. 6 is a flowchart showing the operation of the speech synthesizer 1 of the present invention. (First embodiment, second embodiment) FIG. 7 is a block diagram showing the configuration of the speech synthesizer 1 of the present invention. (Second Embodiment) FIG. 8 shows a user dictionary database 54 with a speech synthesis category of the speech synthesizer 1 of the present invention. (Second Embodiment) FIG. 9 shows a user dictionary database 64 with speech synthesis co-occurrence information of the speech synthesizer 1 of the present invention. (Second Embodiment) FIG. 10 is a flowchart showing the operation of the speech synthesizer 1 of the present invention. (Second Embodiment)

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Speech synthesizer 11 Document database 12 Sentence extraction part 21 Speech synthesis engine 22 Speech synthesizer 23 Syntax analysis part 24 Error candidate production | generation part 25 Speaker 26 Speech synthesis main dictionary database 27 Speech synthesis user dictionary database 31 Speech recognition engine 32 Speech recognition part 33 Reading candidate generation unit 34 Accent extraction unit 35 Microphone 36 Speech recognition dictionary database 41 Candidate selection unit 42 Synthetic speech user dictionary registration unit 51 Category registration trigger switch 52 Category extraction unit 53 Category extraction information database 54 User dictionary database with speech synthesis category 61 Co-occurrence information registration trigger switch 62 Co-occurrence information extraction unit 64 User dictionary database with speech synthesis co-occurrence information

Claims (16)

A dictionary database that maps multiple words to multiple readings;
A speech synthesis engine that refers to the dictionary database, generates a reading group from text data representing a group of words, converts the reading group into speech, and outputs the speech from a speaker;
A selection unit for selecting a part corresponding to a different reading from the text data based on a correct reading input from a microphone;
A speech synthesizer comprising: a user dictionary registration unit that registers the portion of the plurality of phrases in the dictionary database in association with the correct reading. The speech synthesis apparatus according to claim 1,
Furthermore,
A voice recognition engine that inputs a user's voice from the microphone as the correct reading and generates a plurality of reading candidates from the user's voice;
The speech synthesis engine generates a plurality of error candidates from a group of phrases represented by the text data;
The plurality of error candidates include a phrase having a plurality of readings, a phrase not registered in the dictionary database,
The selection unit is a speech synthesizer that selects an error candidate corresponding to one of the plurality of reading candidates as the portion from the plurality of error candidates. The speech synthesizer according to claim 1 or 2,
The dictionary database is
A main dictionary database in which the plurality of words and the plurality of readings and a plurality of accents are registered in association with each other;
A user dictionary database,
The speech synthesis engine outputs the speech from a speaker using a pronunciation accent corresponding to the reading group among the plurality of accents,
The said user dictionary registration part is a speech synthesizer which matches and registers the said part of the said several phrases, the said correct reading, and the accent by the said user's voice to the said user dictionary database. The speech synthesizer according to claim 3.
The dictionary database is
It further includes a user dictionary database with categories,
The text data includes information representing a category,
In response to a category registration instruction, the user dictionary registration unit associates the category, the portion of the plurality of words, the correct reading, and the accent by the user's voice, with the category-added user dictionary. A speech synthesizer registered in a database. The speech synthesizer according to claim 3 or 4,
The dictionary database is
It further includes a user dictionary database with co-occurrence information,
In response to the co-occurrence information registration instruction,
The speech recognition engine inputs a connected speech that is a user's speech from the microphone as a connected portion connected to the portion of the text data, and generates a plurality of connected reading candidates from the connected speech,
The speech synthesis engine generates a plurality of co-occurrence candidates corresponding to the connected portion from the text data,
The selection unit selects a co-occurrence candidate corresponding to one of the plurality of linked reading candidates from the plurality of co-occurrence candidates as the linked portion,
The user dictionary registration unit registers the first word, the second word, and the co-occurrence information that associates the first word with the second word in the user dictionary database with co-occurrence information,
The first word includes the portion of the plurality of phrases, the correct reading, and an accent by the user's voice, and the second word includes the connected portion of the plurality of phrases; A speech synthesizer including a reading of the connected portion and an accent by the connected speech. A speech synthesis method applied to a computer having a dictionary database that associates a plurality of phrases with a plurality of readings,
Referring to the dictionary database, generating a reading group from text data representing a group of phrases, converting the reading group into speech, and outputting the speech from a speaker;
A selection step of selecting, from the text data, a portion corresponding to a different reading based on a correct reading input from a microphone;
A speech synthesis method comprising: a user dictionary registration step of registering the portion of the plurality of words / phrases in the dictionary database in association with the correct reading. The speech synthesis method according to claim 6,
Furthermore,
A voice recognition step of inputting a user's voice from the microphone as the correct reading and generating a plurality of reading candidates from the user's voice;
The speech synthesis step further includes a step of generating a plurality of error candidates from a group of phrases represented by the text data,
The plurality of error candidates include a phrase having a plurality of readings, a phrase not registered in the dictionary database,
The speech synthesis method further includes the step of selecting, as the portion, an error candidate corresponding to one of the plurality of reading candidates from the plurality of error candidates. The speech synthesis method according to claim 6 or 7,
The dictionary database is
A main dictionary database in which the plurality of words and the plurality of readings and a plurality of accents are registered in association with each other;
A user dictionary database,
The speech synthesis engine outputs the speech from a speaker using a pronunciation accent corresponding to the reading group among the plurality of accents,
The user dictionary registration step further includes a step of registering the portion of the plurality of phrases, the correct reading, and the accent by the user's voice in the user dictionary database in association with each other. The speech synthesis method according to claim 8.
The dictionary database is
It further includes a user dictionary database with categories,
The text data includes information representing a category,
The user dictionary registration step associates the category, the portion of the plurality of words, the correct reading, and the accent by the user's voice in accordance with a category registration instruction. A speech synthesis method further comprising the step of registering in a database. The speech synthesis method according to claim 8 or 9,
The dictionary database is
It further includes a user dictionary database with co-occurrence information,
In response to the co-occurrence information registration instruction,
The speech recognition step further includes a step of inputting a connected speech that is a user's speech from the microphone as a connected portion connected to the portion of the text data, and generating a plurality of connected reading candidates from the connected speech. ,
The speech synthesis step further includes generating a plurality of co-occurrence candidates corresponding to the connected portion from the text data,
The selecting step further includes a step of selecting a co-occurrence candidate corresponding to one of the plurality of concatenated reading candidates as the concatenated portion from the plurality of co-occurrence candidates.
The user dictionary registration step further includes the step of registering the first word, the second word, and the co-occurrence information that associates the first word with the second word in the user dictionary database with co-occurrence information,
The first word includes the portion of the plurality of phrases, the correct reading, and an accent by the user's voice, and the second word includes the connected portion of the plurality of phrases; A speech synthesis method including reading of the connected portion and an accent by the connected speech.   A computer program for causing a computer to execute each step of the speech synthesis method according to claim 6. A dictionary database that maps a plurality of words to a plurality of readings, and a storage unit that stores a computer program;
A control unit for executing the computer program;
Speakers,
With a microphone,
The controller is
A speech synthesis engine that refers to the dictionary database, generates a reading group from text data representing a group of phrases, converts the reading group into speech, and outputs the speech from the speaker;
A selection unit that selects, from the text data, a part corresponding to a different reading based on a correct reading input from the microphone;
A computer comprising: a user dictionary registration unit that registers the part of the plurality of words in the dictionary database in association with the correct reading. The computer of claim 12, wherein
Furthermore,
A voice recognition engine that inputs a user's voice from the microphone as the correct reading and generates a plurality of reading candidates from the user's voice;
The speech synthesis engine generates a plurality of error candidates from a group of phrases represented by the text data;
The plurality of error candidates include a phrase having a plurality of readings, a phrase not registered in the dictionary database,
The selection unit is a computer that selects, as the part, an error candidate corresponding to one of the plurality of reading candidates from the plurality of error candidates. The computer according to claim 12 or 13,
The dictionary database is
A main dictionary database in which the plurality of words and the plurality of readings and a plurality of accents are registered in association with each other;
A user dictionary database,
The speech synthesis engine outputs the speech from a speaker using a pronunciation accent corresponding to the reading group among the plurality of accents,
The user dictionary registration unit is a computer that registers the portion of the plurality of words, the correct reading, and the accent by the voice of the user in the user dictionary database in association with each other. The computer according to claim 14.
The dictionary database is
It further includes a user dictionary database with categories,
The text data includes information representing a category,
In response to a category registration instruction, the user dictionary registration unit associates the category, the portion of the plurality of words, the correct reading, and the accent by the user's voice, with the category-added user dictionary. The computer to register with the database. The computer according to claim 14 or 15,
The dictionary database is
It further includes a user dictionary database with co-occurrence information,
In response to the co-occurrence information registration instruction,
The speech recognition engine inputs a connected speech that is a user's speech from the microphone as a connected portion connected to the portion of the text data, and generates a plurality of connected reading candidates from the connected speech,
The speech synthesis engine generates a plurality of co-occurrence candidates corresponding to the connected portion from the text data,
The selection unit selects a co-occurrence candidate corresponding to one of the plurality of linked reading candidates from the plurality of co-occurrence candidates as the linked portion,
The user dictionary registration unit registers the first word, the second word, and the co-occurrence information that associates the first word with the second word in the user dictionary database with co-occurrence information,
The first word includes the portion of the plurality of phrases, the correct reading, and an accent by the user's voice, and the second word includes the connected portion of the plurality of phrases; A computer including a reading of the connected portion and an accent by the connected voice.

Images